The error structure of SP models is a somewhat neglected topic. Utility functions from SP data cannot usually be used directly in demand forecasting. Market segment techniques should be used to deal with this till new techniques have been developed. More research is needed.
Revealed preference and stated preference data have traditionally been analysed by different methods. The authors show the value of SP. A combination of both methods would be a very useful framework for empirical studies.
Guidance is given on designing SP experiments. A good design will result by trial and error from gradual amendment of earlier drafts. Allowance must be made for variations in taste, for example in valuations of time and cost.
